Kubernetes

Your network's DNS,
and your cluster's too.

TelltaleDNS runs from one Helm chart. Your devices get filtered, encrypted DNS on one address, your pods get named by the workload that runs them, and a Raspberry Pi can join as the second DNS server, all managed from one place.

Devices on the network and pods in the cluster both ask TelltaleDNS. Inside Kubernetes, a controller and resolver pods share one address; a Pi outside joins the same cluster as the second DNS server. Cluster names go on to CoreDNS. 📱 Devices 🍓 Pisecond DNS ☸️ Kubernetes One addressDNS · web UI · cluster controllerUI, API, lists, log resolver resolver Your podsmedia/deploy/jellyfin … CoreDNScluster.local one configuration, one query log, managed from any node (illustration)

What you get

Built for running in a cluster

One Helm chart

Published to GitHub's registry on every build, for amd64 and arm64, with a values schema that rejects a mistyped option by its path. Settings and Secrets come from values, so Argo CD and Flux manage it like anything else.

Real client addresses

Per-device statistics and rules need the real sender. The DNS Service keeps it, and if something in between hides it, the web UI says so.

One IP for everything

DNS, the web UI, and the cluster port can share the DNS Service's address: one IP to hand out in DHCP and to remember.

Resolver pods that scale

Stateless resolver pods join the controller by themselves, take its configuration and compiled lists, and ship their query log back. A pod leaves the Service within about 4 seconds of failing, and a rollout drains it first.

A Pi in the same cluster

A Raspberry Pi outside Kubernetes joins as a full member and second DNS server: the same lists, groups, and rules, its queries in the same log, and all of it managed from either side.

Pods as devices

Pods are named by the workload that runs them, so a Deployment is one device across every rollout, with its own top domains, timeline, and anomaly baseline. Namespaces get groups with a label.

Failover you've tested

A drill takes a node out for a few minutes and reports which devices moved to the other server, how fast, and which ones are stuck. On a resolver pod it also measures how long the Service kept sending to it.

Changes that canary

Each configuration change can reach the resolver pods first and the Pi after a bake time, and the cluster goes back to the last good version by itself if the canaries start failing.

Monitoring included

A ServiceMonitor, alert rules, and the Grafana dashboard from chart values, plus a NetworkPolicy and DoT, DoH, and DoQ with a cert-manager certificate.

Agent-ready

Workloads are in the API and in MCP (list_workloads, get_workload), and a change an agent proposes is a plan you approve.

Your cluster's own DNS

Point the cluster at it

With pod naming on, TelltaleDNS answers your pods too: they're filtered by their namespace's group, show up by workload, and still resolve Service names, because TelltaleDNS sends cluster.local on to CoreDNS. Try it on one workload, then make it the cluster's default.

  1. Turn on pod naming in the chart values, with CoreDNS's address and a fixed address for the DNS Service.
  2. Try one workload with dnsPolicy: None and TelltaleDNS as its nameserver.
  3. Make it the default with the kubelet's cluster DNS (one line in k3s's config file), with CoreDNS second in case TelltaleDNS is unreachable.

The step-by-step guide →

See it

Workloads, not pod addresses

The Clients page's Workloads tab: each workload by namespace and kind, its group and where the group came from, its pods, and its queries and blocks in the last 24 hours
The Workloads tab (a kind cluster with made-up apps)
Cluster page: a live topology of a Kubernetes site and a Pi
The Cluster page draws the sites, pods by Kubernetes node, and each node's share of the queries (example data)

Start small. mode: allInOne (one pod) answers tens of thousands of queries a second. Move to mode: scaled or add a Pi when you want a second server, and turn on pod naming when you want to see your workloads. Every step is a chart value.